Backup restore testing

Prove the recovery path before production.

A backup file proves that something was written. Restore Rehearsal tests whether a reviewed backup can restore a disposable application, survive a candidate upgrade, and produce useful evidence.

The rehearsal

Restore, check, upgrade, check again.

The recipe is data, not an inferred promise. Rehearse validates it before the drill and keeps the execution inside a project-scoped disposable Compose environment.

  1. Validate the declared recipe.Reject unsupported or unsafe configuration before starting the stack.
  2. Create an isolated Compose project.Use a unique project name and dedicated temporary workspace.
  3. Restore the declared backup.Exercise the project’s reviewed recovery command rather than assuming the backup is usable.
  4. Run the pre-upgrade checks.Confirm the restored baseline is healthy enough to continue.
  5. Apply the candidate upgrade.Test the declared target and any required intermediate versions.
  6. Run post-upgrade checks and retain evidence.Write an allowlisted report, then remove the disposable environment.

Synthetic first run

Validate before executing.

The repository includes a synthetic example without real credentials or private backup data.

rehearse validate examples/minimal/rehearse.yaml
rehearse run examples/minimal/rehearse.yaml --report rehearsal-report.json
Review required: never point a first run at production data. Inspect the recipe, images, commands, backup source, checks, pull policy, and cleanup behavior before execution.

Where it helps

Built for stateful self-hosted and Compose projects.

Good fit

  • A stateful application already has a Docker Compose deployment.
  • Backups exist, but realistic restore checks are manual or missing.
  • Database migrations or skipped releases create upgrade uncertainty.
  • A team wants repeatable evidence before a human release decision.

Not a replacement

  • It does not create backups or choose a retention policy.
  • It is not a production orchestrator or deployment platform.
  • It is not a complete disaster-recovery program.
  • It cannot certify compliance, security, or business continuity.

Share-safe evidence

Report status, not secrets.

The versioned JSON report schema contains allowlisted status and timing evidence rather than backup contents or arbitrary command output. Review every artifact before sharing it outside its original destination and scope.

Honest boundary

One control among many.

A successful drill increases confidence in the tested path at that time. It does not guarantee every failure mode, production dataset, infrastructure dependency, recovery objective, or future version.

Read the complete safety model →